2.1

Backgrounds of ADFR

Figure 1 (a) shows an example of the protocol stack of the FSO communication system. This figure is just for the explanation, and each technologies here are not necessary common to the present systems. Users are connected with the lowest layer of FSO, which mainly consists of antennas and laser beams. The source of the laser beams are generated in the transponders, whose technologies are assumed, in this study, to be based on the terrestrial optical fiber communication systems using 1.55μm wavelength band.

Figure 1.

(a) Protocol stacks of the normal FSO communication systems, (b) Protocol stacks of the FSO systems with ADFR (test equipment) explained in 2.3. ADFR test equipment is designed to be inserted between User and Transponder as a simple add-on, and not to influence any versatilities of them.

One of the features of the FSO communication systems with the beam transverse in the air around the earth surface is that this FSO layer is unstable in two meanings: the existence of the power fluctuation to the receiver due to the atmospheric turbulence and the quick and wide change of its statistical property even within one system. On the other hand, technologies developed and used in the high-speed optical fiber communication systems are not designed with these kinds of instabilities, and generally, the flexibilities of the operational points in each devices or components become smaller as the expectations on the FSO communication systems are placed on the superior performances in signal speed and receiver sensitivity. The aim of the introduction of the ADFR is that, by introducing flexibleness in the upper layers to cope with link instabilities, it just makes transponders concentrate on showing their intrinsic capabilities by fixing their operational points.

2.2

Adaptive Distributed Frame Repetition

In the ADFR at the transmitter side, the signal to be transmitted is once split into small units, replicas are generated for each units, and they are transmitted to the free space with the appropriate time intervals. At the receiver side, unwanted replicas that are not needed to be forwarded are simply discarded. The statistical properties of the fluctuation of the received powers have two major parameters of fading ratio (FR) and average of fading duration (AoFD). The number and interval of retransmission in the ADFR are determined mainly corresponding to the FR and AoFD, respectively. The proper ADFR settings would decrease the probabilities of the failure due to fading. On the other side, since the transponders operate at the fixed speed, ADFR does not improve the receiver sensitivity, and the bandwidth will be reduced inversely proportional to the numbers of the retransmission.

2.3

ADFR test equipment

The roles and features of the ADFR test equipment developed in this time are explained with fig. 1 (b). It is simply inserted between User and Transponder, and was designed for them not to lose their versatility and interoperability based on the present standardization. The unit of the retransmission is the frame of Ethernet carrying the IP packet. So, the FR is equal to the frame error ratio (FER). More essential performance indicators of packet error ratio (PER) under UDP (User Datagram Protocol) and throughput under TCP can be measured with the test tools implemented in this test equipment at any speed below 10 Gbps. Functions of ADFR are implemented on the commercially available Linux PCs with 10 Gbps Ethernet interfaces. Numbers and interval of retransmission can be varied from 0 to 9 and from 1ms to 100ms, respectively.

There are three essential functions for ADFR to operate in the real systems: (1) generation of replica frames and alignment of them in order as specified, (2) determination of the appropriate next ADFR setting with some monitoring method, (3) seamless change of ADFR setting without system interruption. ADFR test equipment has functions of (1) and (3), and function of (2) is still under investigation.

3.1

Experimental setup

Figure 2 shows the experimental setup used in the following studies. As the purpose of the experiments included performance tests of the developed equipment itself, the setup was very simplified so that the FSO layer as well as the transponders were removed from the configuration shown in fig. 1(b). Two of the ADFR test equipment were directly connected with optical fiber and 10GBASE-ER interface modules, and one of the two directions had an input port for the optical noise source. Frame errors (or Link loss) occurred during the optical switch placed between optical nose source and input port being on.

Figure 2.

Experimental setup used in the evaluations of PER reduction with UDP traffics and throughputs of TCP traffics. Delay elements to emulate long RTTs are not described in this figure. Fade was generated by adding modulated optical noise on one direction of the signal light between 10GBASE-ER. This low-cost approach however enabled very precise transcription of the varieties of the fade patterns on the signal streams.

Fading patterns that drived the optical switch were made with the random number generator in order to precisely control their statistical properties. So, the generated fading patterns do not necessary reflect on the real atmospheric phenomenon nor the responses to them in the real transponders. Each fading patterns consists of 500k samples, and the sampling interval is designed so that it becomes 20% of the AoFD. So, if the AoFD is 10ms, a fading pattern with 500k samples consumes 1,000 seconds. Figure 3 (a) shows a part of a fading pattern with the conditions of FR: 0.1 and AoFD: 10ms, and (b) shows the probabilities of the durations of the link alive/down with these conditions. According to our basic evaluations with various combinations of FE and AoFD, optimum interval of retransmission in ADFR was about 5 times as long as AoFD, and this value was used in all the following experiments.

Figure 3.

(a) Part of the fade pattern generated with random number generator. Fade ratio (FR) is 0.1 and average of fade duration (AoFD) is 10ms. (b) Statistical properties of the fade pattern in terms of durations of link alive/loss.

3.2

PER improvements to the random fade patterns

We prepared fade patterns with various FRs, and evaluated the capabilities of PER improvements against these fade patterns in terms of the numbers of retransmission. Details of the parameters in this evaluation are summarized in Table 1. Figure 4 shows measured results of the relationships between raw FRs and obtained PERs after ADFR processes. Owing to the proper choices of retransmission interval (5 x AoFD), the measured results were well suited for the design targets. According to the fig. 4, if the requirement to the PER is to be less than 0.001, the four-time additional retransmission relaxes the required FER to be about 0.25.

Figure 4.

Measured results of the capabilities of PER improvements by ADFR. Due to the statistical characteristics of the fade patterns shown in fig. 3 (b), one more additional retransmission makes PER to be reduced accordingly to the raw FER if the retransmission interval was properly selected.

Table 1.

Parameters and results in the evaluations of PER reduction

3.3

Performance of TCP in the long-RTT transmission link

For the FSO communication systems to be versatile, it is preferable for them to be under the reliable data transfer protocols like TCP, however, a long round trip time (RTT) with the use of GEO satellite tends to result in the low TCP throughput which is not worth the physical bandwidth. Among the so-called TCP-boosters that provide superior performances in some specific conditions, TCP-FSO that was used in the following experiments were specially tuned to exhibit performances with the long-RTT transmission links by its unique recovery mechanism from fading [9]. Since ADFR is a different approach to this problem with its PER improvement capabilities, the combined performances are intriguingly to be investigated.

With the experimental setup shown in figure 2, the TCP throughputs on the long-RTT transmission link were measured for two cases (w/o and w/ADFR). Table 2 summarizes the details of the parameters used in the experiments.

Table 2.

Parameters in the evaluations of TCP throughputs in the long-RTT transmission link

The plots with solid line in figure 5(a) shows the measured throughputs of the TCP. A slight FR of 0.01 could be an apparent degradation factor, and the throughputs rapidly degreased as the FR increased. The dashed curve lines were the expected performances when ADFR would be introduced with the numbers of additional retransmission being 1, 4 and 9. These expected performances were the simple calculation results using the original TCP performance (plots with solid line in figure 5(a)), PER improvement capabilities (shown in figure 4) and bandwidth reduction effects both by adoption of ADFR. Since the additional retransmission due to ADFR lost at least half of the original bandwidth, and then the throughput with ADFR became less than 5 Gbps, the area of interest was where the FR was over 0.011. According to the figure 5(a), we can find out some candidates of ADFR settings in this area which would achieve better throughputs compared to the case without ADFR.

Figure 5.

(a) Measured results of the throughput with TCP only, and expected performances with ADFR. (b) Examples of measured results of TCP throughput with ADFR (number of retransmission: 4).

Figure 5(b) shows the examples of the measured TCP throughputs in cases without and with ADFR. Details of the parameters are listed in table 2. The plots of “w/ADFR (Measured results)” mostly followed the expected performances explained with figure 5(a) and outperformed the case without ADFR. This indicates that, if the monitor of the link conditions could continuously and properly determine the ADFR settings, the two technologies of ADFR for PER improvement and TCP for reliable data transport can cooperate to increase the overall performances of the high-speed FSO communication systems on the long-RTT and unstable transmission link.